The present invention relates to an apparatus (SEM) for radiating a charged particle beam onto a sample of a semiconductor device for example and measuring the dimension of a pattern using a signal waveform reflecting the shape of the pattern, such as, a secondary electron signal or a reflected electron signal generated from the sample, and a method using the same.
As described in Japanese Patent Application Laid-Open No. 55-72807, the pattern dimension management in a semiconductor fabrication process uses an SEM that is an exclusively specified scanning electron microscope for semiconductors. The principle of the SEM is to obtain an electron beam image by focusing an electron beam emitted from an electron gun through a focusing lens, scanning a sample image in two dimensionally by a scanning coil, and capturing a secondary electron generated from the sample as the result of the radiation of the electron beam by means of a detector. Since more secondary electrons are generated at the pattern edge portion, the electron beam image at the pattern edge is bright. Therefore, in the SEM, the dimension is obtained by multiplying the distance between edges on the electron beam image by a pixel size.
Although all kinds of methods for automatically detecting the edge position have been suggested, only two of them will be explained hereafter.
(1) Threshold Method
The concept of the threshold method is provided in Patent Literature 1. As shown in FIG. 1, the parts with large signal corresponding to the left and right side pattern edges are called left white band (left WB) and right white band (right WB), respectively. According to the threshold method, a Max value and a Min value are obtained from the left and right WBs, respectively, a threshold level that divides these by a predetermined ratio is calculated, a position where the signal waveform crosses the threshold is detected as an edge position, and the distance between left and right edges is set to a dimension (CD value).
(2) Function Application
The function application is a method that fits a predetermined function with respect to a signal waveform, and sets a point on the function to an edge position. For example, a method using sigmoid function (Equation 1) is explained in J. S. Villarrubia, A. E. Vladar, T. Postek, “A Simulation Study of Repeatability and Bias in the CD-SEM,” Proc. SPIE 5038, pp. 138–149 (2003).
                    y        =                  a          +                                    b              -              a                                      1              +                              ⅇ                                  -                                      c                    ⁡                                          (                                              x                        -                                                  x                          0                                                                    )                                                                                                                              (        1        )            
Here, a, b, c and x0 are parameters, and a value that fits best with the signal waveform is obtained by the least squares method. Sigmoid function is a transformed S curve of which up and low parts are flat as shown in FIG. 2. Since only the WB fits, it is necessary to set a proper fitting region of the signal waveform in calculation of the parameters through the least squares method. x0 corresponds to a position with a median between the maximum and the minimum, and this x0 is typically set to an edge position. The edge position is detected by WB of the left and right sides, respectively, and the distance therebetween is set to a dimension (CD value).
There is a sample that is easily damaged by the irradiation of an electron beam. The typical example is an ArF exposure photoresist which is widely used in the semiconductor exposure recently, and it tends to shrink by the irradiation of an electron beam. In general, the SEM is required of measurement accuracy and a smaller amount of shrink. To reduce the amount of shrink, it is effective to reduce the electron beam energy amount being irradiated, resultantly requiring dimension measurement using a signal waveform with a very low S/N as shown in FIG. 3(b). Moreover, FIG. 3(a) illustrates an electron beam image with a high S/N in a case where the electron beam energy amount is high. Because of this, the primary object is how to accomplish high measurement accuracy with a signal waveform having a low S/N. Also, because in a dimension measuring apparatus the accuracy is expressed in terms of reproducibility, the description hereinafter will use the word “reproducibility”.
Meanwhile, problems in a case where (1) threshold method and (2) function application are applied to a signal waveform with a low S/N is explained below.
(1) Threshold Method
When the threshold method is applied to the signal waveform with a low S/N as shown in FIG. 3(b), it is true that measurement accuracy is not high. Since the reproduction of noise is random, Max value and Min value even for the same sample turned out to be different for every measurement and the threshold level varies. Also, the position where the threshold crosses the signal waveform is deviated by the influence of noise. Deterioration of reproducibility by the results of both sides cannot be solved. Although S/N may be improved by increasing the operator size of smoothing as in FIG. 4, the signal waveform becomes blunt, and therefore the accurate edge result cannot be measured.
(2) Function Application
Compared with the threshold method, the function application is robust against noise. In the threshold method Max value, Min value and the position of a signal waveform crossing the threshold are all influenced by the change in a local signal waveform by noise, whereas in the function application parameters are determined to make a broader region of the signal waveform as shown in FIG. 2 fit overall, so it is difficult to get influenced by the change in a local signal waveform. However, obtaining high reproducibility is limited to a case when the function properly expresses the signal waveform, and in a case where the signal waveform is separated from the function a big error may be caused. In effect, when the inventors conducted experiments for comparing reproducibility of the threshold method with that of the function application based on sigmoid function using diverse samples, it turned out that the reproducibility of the function application was lower than that of the threshold method.
Even though sigmoid function is vertical point symmetry, the signal waveform is not vertical point symmetry. In addition, in case of using the sigmoid function as described above, it is necessary to set a region of the signal waveform to be fitted, and the error in the region set-up may be one of factors that deteriorated the reproducibility. Although the reproducibility may be improved by introducing another function, it is difficult to prepare a function that covers all because the shape of a signal waveform varies over a diverse range.